Seven in ten drugs die during Phase II, and each death burns more than $100M along with two to three years. A startup in Switzerland thinks most of them could be spotted first inside a computer.
Atlas Bio left stealth this week with $12.5M in total: a $10M seed led by Creandum, plus a pre-seed that had not previously been announced, led by First Momentum Ventures and LEA Partners.
What it built is a model trained on some 2 million RNA readouts pulled from human tissue. Those readouts carry the signals that decide when a gene fires and when it stays quiet, and from them the system forecasts how a molecule would act in one patient rather than in an averaged cohort.
Twelve weeks instead of three years
A study that would take years shrinks to a 12-week simulation. Synthetic patients mean one virtual profile can be dosed against many compounds at a single disease stage, allowing comparisons no live trial could run.
Founders Simon Romanski and Sebastian Bayer argue Phase II is a poor filter by construction. Just 30 percent of drugs register an effect on the disease they were aimed at, but of the ones that eventually win approval, some 60 percent turn out to work better against something else. Chemistry is usually fine; the study looked in the wrong place.
Animals as the second target
Washington changed the rules in 2022, ending the statutory demand for animal testing before human trials. Atlas is aiming at the gap that opened. A Basel biotech has already dropped animal studies from a workflow and gone straight to people on the strength of its simulation output.
Atlas counts a top 10 pharma company as a partner and is talking to others. Its founders came out of Google and EPFL.
